CeleMine performance is not up to what many of us expected – I think we were expecting another C300a equals PII450 again. There is no doubt that the castration of the CeleMine’s cache has taken its toll on performance.
In contrast, the C300a (@450) was able to equal a PII450 because the C300a’s on-die cache run at CPU speed while the PII’s off-die cache ran at one-half CPU speed. The net effect was for the overclocked C300a to perform essentially equal to a PII450.
The CeleMine has no such possibility – if is in fact a crippled CuMine. What would be analogous is if the C300a was a PII300 with one cache chip missing off the PCB. I think Intel learned from the Celeron/PII experience and ensured that CeleMines would not be ersatz CuMines.
How well CeleMines compare is a developing issue – as part of this dialogue, the following viewer emails shed some light on what to expect – in particular read the Register article below:
Email from Dean: “The Celeron II is turning out to be a HUGE disappointment. I recently got a 566 Celeron II that easily runs 892 MHz and will do 933 for half an hour or so with a stock small heat sink fan. That is OK; what is NOT OK is the the darn thing just can compute its way out of a wet paper sack. I do a lot of SETI and have a multitude of Celerons, Athlons, P2s and P3s. This chip is the worst money spent of the lot.
The Celeron 366s running at 550 on ASUS P299 boards with slockets and
Win98SE do a SETI unit in anywhere from 7 hours 15 minutes to 8 hours
30. Average is from 7 hours 55 minutes to 8 hours 5 minutes. The
Celeron II, with its identical cache size running at a FSB of 105
showing 892 on WCPUID does a unit in 7 hours 21 minutes to 7 hours 45
minutes. That is far worse per MHz than the Celeron “1”. This thing is nuetered in some other way than just cache disabling. I have a P3 500e at 700 doing units in 4 hours 52 minutes to 5 hours 30 minutes.
This is a show stopper for me. I won’t be buying any more of them at
any price. The 38 dollar Celeron 366s are performing in the same ball park. Now that the P3 prices are rising for a while, I await the new on-die cache AMD stuff. Intel is making a big mistake nuetering them to that degree – they are less capable than the old Celerons. I feel cheated.”
Email from Ty:
“I’m a regular reader of Overclockers.com. I was surprised to read Dean’s email regarding the “worse per MHz” performance of the Celeron II.
I thought that everyone had already figured out by now that SETI@Home has a very large working set (about 14 mb), and the nature of its fast fourier transforms is such that it needs to access all of that set all the time. The bottom line for SETI performance is not CPU speed. Once your CPU is ~400 MHz or better, the determining factor is your FSB.
Both Dean’s 550 MHz Celeron and his 850 MHz Celeron II are running on a 100 MHz FSB. Hence, the scores are approximately the same. His P3 500e @ 700 MHz is running on a 140 MHz FSB. He should expect his time to be in the ballpark of ~8 hours * 100/140, or about 5 1/2 hours. I see nothing unusual or unexpected about the results Dean observes.
The P3s also have a slight speed advantage over both the Celeron and the CeleMine due to the larger and more associative cache. However, it’s really the 140 MHz memory bus that helps his P3 win.”
Email from Scott:
“In regard to Dean’s email, SETI is an extremely cache intensive. If Dean were to buy a spitfire CPU, he would get disappointing results as well. A Pentium3 500 out performs a Pentium3 700 when running SETI.
Go to this article from the Register to find out.”
Email from Gerry:
“I’ve got a Cel2 566 happily running at 900, and while I have not put a stopwatch to my Seti work units, I can say with certainty that my old Cel466@601Mhz took about 11 hours to do a work unit. My Cel2 takes a little over 7 hours. Dean says that a Celeron 366 is “in the same ball park” as a new Cel2; with all respect, I would like to know where the line starts for that kind of 366, I want a couple!!
Also, his margin of performance between the P3 500e@700 and the Celeron2@892 seems inconsistant with web-published comparative benchmarks between Celeron 2 and Pentium 3. He seems to have a much wider performance gap than everyone else. Perhaps there is some sort of optimization in the Seti client that gives this edge to the P3.
Be the first to comment