Back last summer when I first started experimenting with Knoppmyth, the best price/performance ratio was a Barton core Athlon XP 2500+. But between then when I was playing, and now when I'm building the CPU world has changed a great deal. Athlon XP chips are getting harder to find, and with scarcity comes higher prices. I can actually get an Athlon 64 3000 (939) for $45 CDN less than the cheapest remaining XP processor.

In the past I've always avoided the low processing power CPU's (the SX's in 386 & 486 days, the Duron's and Celeron's). The price difference between a Sempron and an XP has me seriously considering one of those this time round.

Maybe I'm missing something, but the only difference I can see is the Barton core XP's have 512 K of L2 cache, and the Sempron's have 256 K. The Knoppmyth suggested specs used to say an XP 1700+ for software encoding, but you could get away with less if you're using a hardware encoder (I've got a PVR-250). Unless I'm mistaken a XP 1700 would have been the Thoroughbred core, which only had 256 K L2 Cache. Is a Sempron really just a slightly reworked Thoroughbred core, a Barton core with less cache or something completely different?

And how much power is enough? I know the Wiki says don't cheep out on the processor. Is 2400 enough, or 2500, or 2600 or 2800? The price of a 28 is about double that of a 24. Above 30 the Semprons turn into Socket 754 processors, which won't fit in my Socket A 7NIF2. (3000+ Semprons are available in both Socket A and 754)

Before someone chimes in with a stuffy, "well you won't be able to encode HDTV with one of those," let me put your mind at ease. I don't have an HD capable TV and probably won't be buying one for a couple of years, until the prices become more "blue collar" affordable. Obviously my encoder card won't do HDTV either. My cable provider wants $400 for a HD DCT, and no they don't rent them, thank you very much. To buy all that new equipment is well out of my price range. Besides, when I'm ready to go HD, a bottom end Athlon 64 will probably be a 3800 to 4000 (more than enough to handle the job), and priced well below $100.

My current expectations of my Knoppmyth box is, at a maximum, record one show while watching something recorded earlier. I suppose it is possible that I might get addicted to this thing, and decide to put in a second tuner (another PVR-250, or 150/500 when drivers become available).

I read every day's postings on the forums in every major grouping except HDTV (though admittedly I do skip a few topics that I don't think apply). And yes, I have searched the forum for the word "Sempron" and a few other CPU speed related terms.

Comments, suggestions and questions are welcome.

Ian

----
Edited to make the following corrections:
- "XP 1700 would have been the Tunderbird core" to Thoroughbred
- "slightly reworked Thunderbird core" to Thoroughbred
Must have been thinking too Mozillish
- "Above 28 the Sempron's turn into Socket 754 processors"
to "Semperons" and "30"

My recommendations have not changed. A 2400+ is more than enough to record while watching one program.

_________________
cesman

When the source is open, the possibilities are endless!

just pick the cheapest cpu you can find that fits your board any of them should handle what you want plus and additional card should you decide to add it. I had a 1700+ for a long time and it handled a pvr250 + a software card and I could record two shows and watch a previously recorded show with no problems. And yes a sempron any of them will be faster than a 1700+.

_________________
Have a question search the forum and have a look at the KnoppMythWiki.

Xsecrets

With my machine I don't have any problem recording one show and watching another. In fact it runs below 3% CPU while doing this and I think "top" takes 1%.

From all of the information in these many sites, it's because of the PVRx50 hardware encoding and decoding. So, if you plan on using one 250s to record I'm betting you can get away with a pretty cheap CPU and just use the processor for the TV-Out. I'm hoping to add a 150 when they start working, but if they don't have hardware decoding I might add a 250 and not worry about it.

PVR-350
2.40 GHz Intel Celeron
128 MB DDR SDRAM
200 GB Maxtor Ultra-Quiet
CD-RW

-Lane

Athlon prices really have gotten weird, since AMD has basically discontinued production of anything slower than the 2800+ Barton. The result is that there are Athlon XP (socket A/462) chips which are more expensive than faster Athlon 64 (socket 939) chips. However there are still plenty of resonably priced Socket A options. With a PVR-250 the main remaining CPU intensive tasks is commercial marking, and frankly it's really doesn't matter if that takes twice as long. (I'm ignoring transcoding but mostly because I figure MPEG2 will let you burn to DVD and avoid that.) As a result any CPU faster than about 1.2Ghz ought to work just fine for you. Some research at my favorite price comparison sites shows that there are still a lot of Palomino and Thoroughbred processors out there at fairly reasonable prices (USD). Prices in CDN will be higher and you might have a harder time finding them, but they're out there.

Athlon XP 1800+ (Palomino) Athlon XP from $44
Atlon XP 2100+ (Thoroughbred) from $54
Athlon XP 2400+ (Thoroughbred-B) from $60

Trying to run a system with less than a 1Ghz processor is doable but right on the edge where things get iffy. As the person who wrote at least one of the wiki pages I think you're talking about, the intent was that by chosing a CPU from the "safe" range you wouldn't have to worry about things like cache and the like. Any 1.8GHz CPU should do the job. Athlon, Duron, Sempron, Pentium, Celeron, ... The reason that the 2400+ and 2500+ Bartons were popular was that they were cheap, run cool, and gave you lots and lots of "headroom". I'd go so far as to say they're probably overkill for recording NTSC with 1 or 2 PVR cards.

BTW - The reason that a 3% CPU usage on a 2500+ while watching live TV doesn't scale to CPU 10x slower has to do with bursty demand and "real time" deadlines when dealing with video streams.

also decoding mpeg2 from a pvr250 without using the hardware decoding of a pvr350 will be far from 3% I would guess it'll be more like 40%, but still plenty of headroom for two cards.

_________________
Have a question search the forum and have a look at the KnoppMythWiki.

Xsecrets

I've got it running right now, with "live" TV about 5-10 minutes behind real time so we're talking capture, shuffling to disk, retrieving from disk, decoding, and playback. (No time lag reads right from the disk buffers and shaves about 1% off this.) Top says somewhere between 98% and 95% idle with a rare excursion to "only" 93% idle. Top can be fooled but ... I'm recording MPEG2 with a PVR-250, and playing it back with the Nvidia MX4 built into the NForce2 IGP, without XvMC, the CPU is not overclocked or anything..

Here is a snapshot of top (I waited for a usage peak where both mythbackend and mythfrontend where in the list):


Here is the CPU info:

BTW - I can push the usage up to 5-8% by calling up the program guide with the TV playback running in a window, but the % idle still never dips below 90%.

wow I never would have believed there was that much difference from the 1700+ cause it always ran 50+ % when decoding mpeg2 from the 250.

_________________
Have a question search the forum and have a look at the KnoppMythWiki.

Xsecrets

Which core does your 1700+ have? The Baron pipeline seems to be bettered tuned than it's predecessors and performs better even at the same clock speed as a result. Not to mention the faster memory bus (333 vs. 266) which probably helps a bit too. Actually speaking of that, what are your memory timings?

Thanks for all the great advice. Looks like my Myth box is going to be a fairly reasonably priced machine once I get a few rebates back. It will consist of (prices in CDN $ - as of this writing $1.00 US = $1.2188 CDN):

101.99 - Chaintek 7NIF2 (after rebate)
149.99 - Hauppauge PVR-250 (on sale at that price about once a month)
129.99 - Western Digital 250 Gig 8 meg cache hard drive (boxing day sale and rebate)
_76.95 - Sempron 2400+
120ish - Haven't picked it yet (512K Mushkin or Kingston CAS 2 or 2.5)
__0.00 - Old case and power supply I've got laying around
40.53 - GST (tax)
------
619.45 - (rebates are in US $ so may very by a few bucks as exchange rates flux)

--------

In my original missive I asked the question:


Well I did a little research on that point and now can answer it myself. (Why oh why do I get these questions in my head, and then stay up all night reading dozens of websites till I have the answers!)

It turns out that the Semprons are Barton cores with half the cache disabled. That's not as much of a conspiracy as it initially sounds. There are good technical reasons for it.

As most of you know processors start out life together with many other of its brethren on a single silicon wafer. Not all the chips on that wafer work perfectly at the target speed for that batch. Some don't work at all due to contaminants falling on some part of the wafer. It is common practice in the industry to take a chip that failed to run stably say as a 3000+, and retest it as a lower end processor. If they find it stable at a slower speed, it is just rebined as that speed of chip. (Makes you wonder if some of the real successful overclocking stories are chips that were initially targeted at a higher speed to begin with.)

A similar situation happens with the Barton/Sempron cores. Cache takes up silicon real estate. With the increase in cache size from 256 on the Thorton to 512 on the Barton, (and many other changes that made the Bartons run cooler and faster) the physical die size got larger. That means fewer chips per wafer. That larger die also means a greater chance for some contaminant to ruin the chip and further reduce the yield. Yields on the Barton line were lower than had been expected.

But what if the actual guts of the processor are fine, and the contaminant only ruined part of that large space occupied by the cache? You could garbage the chip because it didn't meet the specs of what you were trying to produce. Or you could disable the bad part of the cache, with a simple laser cut of a trace and re-bin it as a processor with a smaller cache. Then sell what otherwise would have been garbage at a discounted price and make some cash from the problem with the cache. (Couldn't resist that play on words.)

And besides, the Duron line was getting kind of long in the tooth, and stale. Think of it. You could repurpose the production line that used to make Durons, still have a discount chip, and have that cheap chip be more powerful than the old Durons. And if the demand for this crippled chip becomes high, or suddenly you solve the problem and Barton yields go up, or marketing/technology forces cause a newer high end chip into being produced, just keep producing regular Bartons and cut the trace on ones without flaws.

And hence the Semprons came into being.

Ian

The situation is more complex than that. Usually the maker knows how many of the higher speed chips they want/need to meet demand during a particular period. They test until they get that many and then don't bother testing the rest at the higher speed, this cascades on down the line. Thus many of the chips spec'ed for a lower speed would actually pass at a higher one, they were just surplus to the demand.