Overclocked Celerons. [was:Re: seti@home]

Eugene Leitl (eugene.leitl@lrz.uni-muenchen.de)
Mon, 12 Jul 1999 22:54:18 -0700 (PDT)

dan writes:

> I'm not a real overclocker: those folks do crazy things like immersing
> the entire motherboard in mineral oil and using refrigeration
> equipment to cool the oil. I merely took advantage of Intel's mistake

I wonder why nobody uses flurocarbons -- mineral oil is so messy. It's not very cheap, ok.

I wouldn't use overclocking for business-critical/scientific applications, because timing bugs are unpredictable, and thus hard to debug. For gaming and office uses they are certainly great, though.

> with the Celeron 300A, certain of which are really capable of 466Mhz
> operation. Check out some of the weird overclocking stuff on slashdot.

Thanks for providing hard data on overclocked dual-Celerons. It is really strange that few people know that the low-end prices and unit performances are a politicon. Only high-end is profitable, the low end is sold essentially at raw silicon prices (ocassionally even below). It will be crucial for AMD to establish Athlon in the high end price segment, because the number of sold units alone won't save them.

> CPUs are not the only chunk of hardware caught in the inexorable
> march of Moore's Law. Disk drives are still keeping up, which
> never ceases to amaze me. Disk drive have motors! What does that
> have to do with Moore's Law?!! Nothing, but lithographic and related
> technologies have driven up the magnetic recording density at roughly
> the
> same rate as the DRAM bit density. Here's an interesting result:

We're nearing the end of the usability of the spinning bit, whether trick (SNOM et al) assisted, or not. Solid-state bits will win eventually because the substrate preparation costs will soar as bits will shrink further -- while concurrently error rates will rise. It should be possible to utilize almost the entire 200/300 mm wafer (95%..99%) surface for nonvolatile storage with next-generation nonvolatile RAM cells which can be smaller than 200 nm. Even not considering access time, this is very hard to beat that with rotating bits on a platter.

> the retail cost of hard disk storage has fallen below $10.00 (US) per
> gigabyte. This means that the cost of storing an MP3-compressed CD
> (50 megabytes for the typical 50-minute audio CD) is now $.50 US. To
> calibrate this, The database at www.cddb.com knows about more than
> 325,000 unique audio CDs. This must be a high percentage of all the
> CDs aver released. An MP3 copy of the entire collection (about 16
> Terabytes)
> would require $160,000 worth of disk this year, and will require
> $16,000 of disk in three years. It would currently fit in about
> 8 boxes the size of a desktop PC enclosure, and will fit in one
> such box in three years. As a point of reference, Microsoft claims
> their 1-terabyte geological image database is the biggest commercial
> database in the world.

Beowulf-type you can easily put 100 GByte worth of EIDE drives into 1U 19" rackmount enclosure, including the PC motherboard with FastEthernet onboard. Even with full redundancy, a massively parallel 1 TByte disk server from such commodity components can very comfortably fit into a half-height 19" rack. At least twice that much with good thermal engineering and (more expensive) higher-density drives.