Discussion in 'Recording' started by angrynote, Feb 16, 2002.
[ January 07, 2004, 03:02 PM: Message edited by: angrynote ]
i don't have first-hand experience with this board, but am in the same boat as you trying to build the perfect system & have been doing a bit of research. read somewhere lately of a "pci throughput" bug associated with the 850 (and 860) chipsets. i don't get the impression this is disasterous problem, but the solution is due to be out later in Q1. i'd love to have made my purhcase yesterday, but until i find out more, i'm probably going to sit on the sidelines.
Anyone have more thoughts/info on this? also, is this bug found on the 845D chipset?
Hey guys - this one's been bugging me too - the throughput bug in 850-860 according to intel's "errata", causes dropped info at speeds over 100 MB/sec, if I read it correctly - This is supposedly NOT the case with 845D chipsets, which seems to be borne out by the fact that I have not been able to find a MoBo with 850 chipset that supports ATA133 Raid - I don't even remember finding one with ATA133 period. If you go to Pricewatch.com and search for "845D raid", you can find a few MoBo's, some with ATA133 - if you do the same search with i850, they all come up as ATA100. My conclusion so far is this: you have to decide whether absolute maximum track count is your priority (845D, ata133 raid) or if you're going to want max plug-ins with a pretty high track count. My impression so far is that DDR is only SLIGHTLY slower than Rambus, and 'way cheaper. SDRAM is a lot slower, and I would only recommend it to someone on a really tight budget.
(Dead Link Removed)
The above thread has biased me toward a DDR MoBo with ATA133 Raid, since I plan to "cast the system in concrete" for at least 2 years and try to actually make noise instead of screw with computers.
Another factor for me is this: The built-on raid chips are NOT created equal - I downloaded manuals for Abit and Gigabyte boards with Raid - The Gigabytes use Promise (whose PCI cards have gotten good press) but the "LE" version which is embedded in MoBo's (at least on Gigabyte) DOES NOT support raid 0+1, only 0 or 1. The Highpoint embedded controller used on the Abit boards DOES support 0, 1, and 0+1. Both embedded chips allow you to use the extra 2 channels of IDE as straight IDE, so you can use 4 HDD's and 4 slow IDE devices (CD-rom, DVD) or 6 and 2, or go with raid. Setting up the second two channels thru Dataport plugin hardware would allow using that as "customer" drive option, as one example.
Another thing - I read (don't remember where, probably MaximumPC) that DDR is expected to become a bottleneck when procs go beyond about 2.4 gHz, so If you plan to "keep up with the Joneses", maybe i850/60 would make more sense - but only if/when they get the 100 MB/Sec thing ironed out.
I'm pretty much decided on DDR/ATA133/On-board Raid, so the MoBo will most likely be the Abit BD-7 Raid, at about $150. I wouldn't think of buying a non-retail box version of a MoBo, living in "stickville USA" - the extra $10 or so would be eaten by the first thing I needed that didn't come with the OEM version.
Hope I didn't just add to the confusion, there's certainly enough of that to go around... Steve
thanks, steve. yeh, plenty of confusion over here, i'm certain of that. interesting, i hadn't heard abt the ddr bottleneck at proc speeds abv 2.4 ghz ... i'm holding out until i find word whether intel plans to release a ddr chipset that supports ddr333- it will close the gap agst the 850 rambus flavor, already proven by sis (the sis 645 chipset has recvd good marks, but i'm not sure on sis' stability -many say intel chipsets will do much better with disk drive intensive apps -ie audio-).
other question i have is how much, if at all, do on board raid controllers drain from other (pci) resources? of pro's i've consulted abt specs for my room, some are pro the likes of a promise controller- while others say it's not worth the extra strain it puts on the system. thoughts on that?
angrynote, knightfly and all others in our shoes, check out this article just released on tom's hardware - on rambus, ddr and the future of the p4.
here's the link to the article, or you can just go to http://www.tomshardware.com
tom's hardware: Rambus - Take 2
enjoy. still have no idea what i'm gonna do with myself.....
A note on "perspective"...
We all know that technology advances like a snowball, growing exponentially faster. So we all must ask ourselves, "At what point do I draw the line?". Do I upgrade now, with what's currently on the market, or do I hold-out for the next bigger (better) thing to come along? At what point do you make your decision to drop your hard-earned cash on the latest technology, knowing that very soon technology will leave it behind? If you are in here reading this, I'm quite sure you understand what I'm saying, and can relate.
I bought my Mac G4-400agp two years ago. It's a dinosaur now. As time goes on, the snowball effect will keep making dinosaurs out of our computers and other electronic gear faster and faster.
Another prime consideration for all of us is cost. The newest technology is always expensive. For example, when a CPU or motherboard has been on the market a while, the price drops considerably as other products enter the market which surpass it. But, if you hold-out for a while for the price to drop, you get the temptation to buy a newer item!
Bottom line ... if you must have it now, buy what will fill your needs, but shop for the best deals. If you don't really need it now, then hold off, because the price on that same item will drop or new technology will come along that will keep you further ahead of the snowball.
SonOfSmawg, great point. you've got to be a millianaire to keep up with the jones'. each person has to assess at what point are most of their pc criteria hit in current technology and bite the bullet.
for me, i'm moving from an all-analog (well did bits n pieces -never more than 10 tracks- on location with a laptop) and this will be my first dedicated pc for audio. i was ready to pull the plug until i heard abt this little intel 850 "bug". with the fix coming down (so i hear) in april plus a 533 mhz bus, i'm crinching my teeth but will wait.
Angrynote - funny you mention the minidisc config. it seems just a couple years ago, i thot mastering down to one of those units yielded me a technological pioneer! haha how wrong.
as far as experimentation goes, i'm game for that, but if these new chipsets will handle 533 mhz, does that mean p4 processors will also be increased (therefore raising the experimental cost to almost a full 2nd computer) to the same speed? i would think they do get a boost.
Hey Mitch (and anybody else considering raid) Here is a comment from RME's FAQ site -
"Promise Fasttrack RAID Controller
The Promise Fasttrack Controller causes performance problems when using the Hammerfall. An operation at lower latencies generates clicks and pops.
Comment: To achieve the highest rating in benchmark tests the Fasttrack controller hogs the PCI bus, thus impacts the data transfer of other PCI devices (the Hammerfall). This is true for both the PCI-card and the on-board version (as on the MSI K7T266 Pro), but not true for the majority of 'normal' Promise RAID controllers.
Promise offer a software tool to set up the controller. Move the slider labelled PCI Bus Utilization from 100% (right side, default) to about 75%. This solves the problems."
This is the comment that caused me to consider just using the raid controller (built-in, in my case) as another pair of IDE channels. Guess I'll havta try it myself and compare. From the sound of things, it wouldn't make any difference whether the raid controller was onboard or add-in - Also, if you throttle back the raid performance in order to get rid of the PCI hogging, then you may as well just use single drives one at a time. I say this because even a raid 0 is apparently only a little faster than a single ATA 133 drive - Also, I haven't seen an add-on Promise card that does ATA133, only ATA100, whereas Soyo's new SY-P4I gives you Hi-Point ATA 133 raid usable as 0, 1, 0+1, or 2 more IDE channels, 6 PCI, AGP Pro, 6 channel audio, 4 usb (1, not 2) 2 firewire, 10-100 Lan, and a SmartCard Reader header. The only downside is only 2 DDR sluts, so you have to live with 1gB (or smother your rich uncle before he can change his will) The other reason I have for staying with the 845D chipset is that I need the machine soon, and of the two chipsets that support WinXP's extended (24) IRQ's, the i850 screws up high speed ATA by corrupting data over 100 MB/Sec. I don't yet know if the (still vaporware) i850E chipset will fix the PCI bandwidth limit or not. I'm not willing to wait - So, for me (and I know who I are) it comes down to the i845D MoBo with the most goodies, which right now appears to be the Soyo SY-P4I "Fire Dragon" - I'll know for sure if and when I ever get their website to let me and this bucket of slug pus known as IE 5.0 finish downloading the manual.
I'm glad to see all the rest of you are at least as corn-fused and pained by all this as I am, 'cause I'd hate to be a hog about it when it's so much more fun to share... Steve
thanks knightfly. i haven't really gone out of my way to obtain raid capabilities. a guy i had been speaking with at steinberg recommended the promise card ... on a subsequent conversation, his view altered that they were better pci hogs, causing some of the inefficiencies you note - and rme now notes this on their site. to be honest, i'm not sure (and am curious!) which has the advantage: pci raid or raid built in to the mobo. even though i think i'm holding out for the next intel 850 board, i'd be curious to find out as rme now recommends this gigabyte board as their standard reference board (has both ata 100 and ata 133 raid on it - if i've read it correctly).
the main reason i wanted raid was to guarentee that i could get 40 tracks at least using 24/96 resolution. i could definately get this track count at 24/48 - but at the higher resolution, nobody yet that i have asked has told me i'd get the desired results.
Hey Mitch - I've been wondering about the real world track counts at various sustained thru-puts too - my old DAW has an 18 gig U2SCSI 7200rpm drive, that has never benchmarked over 15 MB/Sec. in its life - I have a couple of Samplitude projects that were saved in 32 bit float mode, that have 18 or 19 STEREO tracks. These were, as I said, 32 bit WAV files at 44.1 k sample rate. If you do the math, 1 24/96 track takes roughly 3.2 times the storage as 1 16/44.1 track. 16/44 requires 5 MB/Minute, or .0833 MB/Sec. , times 3.2 = .2666 MB/track/second. Multiplying by 40 tracks, you get a requirement of 10.666 MB/second for 40 tracks at 24/96. Given that most upper end DAW programs actually work and save at 32 bit float until you dither down for final product, you would still only need a real sustained thruput from disk of 14.222 MB/Sec. for 40 (mono) tracks at 32 bit/96 kHz - All the later ATA100 and up drives test out at 'way more than that (unless someone's lying) In fact, I vaguely remember numbers in the range of 30 to 35 MB/sec. for ATA 100 drives (7200 rpm, of course) A sustained thruput of 29 MB.sec, if it's real, would theoretically let you have 80 tracks at 32 bit 96 kHz. So, the trick now is, build something, try it, and find out how bad the old "YMMV" bites you.
In fact, I'm ordering all the stuff for my new DAW tomorrow, and one of the first tests I'm going to run once it's together and debugged, is to rip an audio track from a CD, convert it to 32 float @ 96k, copy it to new tracks (total 48 STEREO pairs), change each track a little, eq, flange, chorus, etc., then If I can't choke the 'puter with that, add some more until I know just where the line is. That should at least tell me how close theory and reality are.
As low as the thruput calculates for 40 tracks even at 32 bit, I'm starting to think that possibly the main reason nobody will put their foot in their mouth is either that they don't know the joys of toe-jam, or they haven't a clue because they haven't done it themselves. Another reason for non-commital could be that they know how much difference proper drive maintenance makes, but they DON'T know how anal you are about de-frag/scandisk routines. I know that the one time I let defrag go (same 32 bit files) I thought my drive was going to self-destruct ! When I finally figured out why it sounded like a Maytag full of marbles, it was like, "Doh!"
My philosophy on built-in raid is this: It costs less than add-on, you can use any of them as 2 more IDE channels if you don't want raid, or you can dis-able the raid altogether in BIOS. So far, the only thing missing from the Promise version of built-in raid is mode 0+1. This has led me to only look at boards with the HiPoint raid controller onboard, which include Abit and Soyo, but NOT GigaByte. The thing that makes me nervous about the 850 chipset is this: If it corrupts data at ABOVE 100 MB/sec, why not at 99? 97? How do we know it's OK at the rate we're working at? Intel's site makes no mention of the problem existing on i845D chipsets, so I have decided to take the 10% or so hit on DSP in lieu of worrying about lost data on recordings. If I were (a) more trusting and (2)less in a hurry, I might wait to see if the i850E is the fix for their corruption problem or just more smoke and mirrors - since I'm neither of these, I'm going with an 845D raid MoBo with 1 GB ram, 2 gig Northwood (just in case the MoBo can't overclock much), 6 80 GB 7200 rpm Maxtors, Matrox G550 32mb DDR Video, 600 watt PC Power/Cooling PS, at least a 10-bay tower, with CD-RW and DVD-RW firewire drives. I intend to put the PC out of the room in a special cooled and filtered enclosure, so the firewire will allow the optical drives to be in the control room. The other two drive bays in the tower will be Dataports for customer drives. Since I will be using firewire for optical, I can use all the IDE channels for HDD's. Now that I've almost committed, watch the dirty $#^$&%%'s solve all known "issue"s with the i850E day after tomorrow... Steve
that's alot of info, steve. i will be waiting on pins and needles to hear the outcome of your hard drive tests. and those calculations you made in the top portion of your thesis were assuming just ONE high-end drive? that's not bad, actually. better than i thought. well, fortunately i guess, i'm not one of those in a hurry, as i'm still working on the construction of my space, i have a little more time to wait and see how large a rabbit intel can pull out of their hat. i thought the article on tom's hardware re: rambus was pretty interesting - so i'll give them the opportunity to show us a fix before i likely to the same route as you (altho hopefully by then intel 845 will be able to handle ddr333).
good point on the built-in raid that you can just turn it off via bios if you don't like it. this is going to sound like a silly question, but i've been looking at the western digital special edition jb drives (jumbo buffer of 8mb) - where do i look in this mess of numbers to find the sustained throughput ?
I know we're getting off the subject of this thread, and I know it's my fault (as usual) so, OPUS, move it if you want - I think this is important (but hell, ALL my stuff's important, some of it's even correct...)
This just in - cancel all euphoria caused by the above wishful thinking - Just got thru reading an article in this month's MaximumPC entitled "raid revisited" - These boys are nothing if not masochists... The previous article they ran on raid was apparently tainted by their benchmarking software, so they came up with their own test. They took a machine with 768 MB ram, installed 4 WD 1200 drives (Western Digital WD1200BB WD Caviar 120GB ATA 100 7200RPM), an Adaptec 2400A raid controller (2400A RAID KIT ATA/100/66 4CH 32BIT PCI 0/1/0+1/5/JBOD) (note that neither the drives or the controller are ATA 133)
Just in case anyone reading this isn't sure what "ATA 100" means, it is (supposedly) the fastest transfer rate the unit is capable of in burst mode, measured in MB/Second.
Compare to this scenario: A car salesman tells you his 550 horsepower sports car does 0-60 in 2.8 seconds, and can get UP TO 40 miles per gallon - What this means, is that with $600 tires (not included) the accelleration is possible with a Professional driver on new, dry blacktop, using BX-10 traction compound, with a good stiff tailwind, etc - and if you hang the car out over a 1000-foot drop with the engine idling, drop the car and measure how many molecules of gas the engine used up before it hit the ground, then convert molecules of gas and 1000 feet to miles, and you have 40 mpg. (The car will NEVER get MORE than that, and that's the point of this too-long story.) Rambling over - there will be a test on the next part...
The boys at MaxPC then set up a 540 MB RamDisk and proceeded to time the transfer (both ways) of "small, medium and large files", using NTFS in WinXP, and with cluster size set to match stripe size in various raid settings. Their best disk write time was 35.00 seconds using Raid 0 x 4, their best disk read time was 19.00 seconds .
A single disk test turned up 42.00 seconds write, 28.67 seconds read. Since few of us will be trying to RECORD 48 tracks at once on a single computer, I will limit the number stuff to READ performance converted to tracks at various sample rates and bit depths. To stay within the margins, I have rounded 44.1 off to 44 (for example) ONLY in the title - I used 5 MB/Min for 16 bit 44.1 mono tracks, converted to MB/sec., then re-calculated for other depth/rate settings, and finally rounded DOWN to the nearest # of tracks. Keep in mind that any SERIOUS software will allow you to do DSP and storage at 32 bit depth, and this should be used whenever possible unless you can't stand the overhead and won't be doing ANY DSP, but only basic editing. 32 bit float mode, while not ideal, is the best you can find in any reasonably priced DAW software and will help you "throw out the least amount of baby with the bathwater" when burning to final product. What this means (to me, anyway) is plan your drive track count around 32/96, unless you're just in a pissing contest about whose stuff can do the most tracks... The following table shows first the probable track count (playback) of a single WD 1200 drive at 18.83 MB/Sec, then in the next column is the track count under a 4-drive raid 0 setup at 28.4MB/Sec. All tests were done using NTFS file system, with cluster size matched to raid stripe size.
Please bear in mind that these are numbers taken from a NON-audio magazine, achieved with ATA 100 components, then converted by me to THEORETICAL track counts - Guess what my response will be to anyone dumb enough to go buy these components and then blame me because "I only got 36 tracks instead of 41..."
DISK READ ESTIMATES - # OF TRACKS AT VARIOUS DEPTH/RATE
Depth/Rate MB/Sec/Trk Single Drive Raid (4 Drv)
16/44 .0833 224 340
24/44 .125 148 226
24/48 .136 137 207
32/44 .167 112 169
32/48 .181 102 156
24/96 .272 68 103
32/96 .363 51 77
Who's to say (unless they've actually done it) what the results would be with a different controller or drives or ATA 133, or even if the exact components would work the same in your machine ? This is mainly just to know what to hope for until reality grinds you and your pathetic attempts into the (oh, wait a minute, that was something I was working on for Stephen King) First grinding - these #'s DO NOT take into account the PCI hogging characteristics of raid controllers - Because of this, I intend to consider myself lucky to get 40 tracks playback at 32/96 on a single drive, but who nose? (not MY nose) Gotta go, full moon must be coming early this month, damn hair always gets caught in the keyboard... Steve
P.Sp - Mitch - As far as I know, you either do your own experiments or rely on guys like Maximum PC to get REAL #'s on drives - Marketing types would like you to believe ATA 100 really means 100 MB.Sec. As you see from the above, BULL...
BTW, what do you have to do to get some damn columns to space out ? I tried tabs first, then 15 spaces between columns - What gives ? Anyway, the order of data is correct, just imagine a whole bunch more spaces between the numbers... Steve
I aint touching this post...just "moderating" the moderation of reading...umm..yeah, what he said!
This is great stuff and it doesn matter what the Topic is..as long as it stays relevant to out addictions..err, I mean hobby, err. I mean work..err..nevermind
With the Ata133 Raid specs out there now..what about the prices and availability of the ATA133 drives?
Also are we getting ourselves into trouble again by trying out new technology?
Like Via's data transfer bug on the KT133a...was there ever a patch? I doubt it since those chipsets are slowly fading away into the past and the graveyard of junk mobo's
Anyhue..the search will always go on for the best system and best components to go with them..
And this is where this wonderful place we call RO comes into play!!
Great post on track counts... one small clarification... ATA66, 100, 133 are just the speeds the *interfaces* are capable of- while a drive manufacturer is happy to not post the transfer rates in big numbers, fact is all the ATA drives cap out (individually) under even 66 Mbps, so with *one drive,* noone would ever notice the difference between the three most recent flavors of ATA. Multiple drives on one cable, though, add up, so going to ATA133 may make sense if you're going to be doing RAID or writing a multitrack project off one drive to a high speed CDRW on the same cable without bouncing your mix to disk first...
Separate names with a comma.