Home |
Search |
Today's Posts |
#81
|
|||
|
|||
Bob Cain wrote:
Scott Dorsey wrote: Bob Cain wrote: wrote: There are performance benefits especially when doing 64bit computations on a 64bit processor. Most certainly, but what's in question is the relevance of that to a DAW. Right now, it's totally irrelevant. But I bet it turns out to make it easier to design good fast reverbs. Double-precision floats can be a good thing for reverb simulation. I may be wrong but I think that FP units have 64 bit wide data paths even in 32 bit machines. I believe you're right. In and of itself, 64 bit is really about integers and addresses. Surprisingly, we're getting to the point where 32 bit addressing is starting to look a little tight. The largest 32 bit integer is about 8 billion, which corresponds to the ever-so-common 8 gigabyte addressing. Right now I'm building customer machines with 0.5 and 1 gigabyte of RAM, so 8 gigabytes of RAM is quite easy to see on the horizon. In fact I'm contemplating building a 2 GB machine today. Building machines with very large amounts of real memory is stimulated by the fact that CPU speed has long been outpacing hard drive speed. If you really want a 3+ GHz machine to exploit its potential processing power, you don't run much directly off the hard drive. Since large amounts of RAM are being used to cache hard drives that keep getting larger and larger, computer real memory size has to in some sense, keep up with increases in the size of the hard drive. Computers have to work with virtual address spaces, and even 8 GB addressing is too small when the address space is being used to work with an entire database. Again, very large databases are facilitated by large hard drives as well being demanded by increased use of graphics and imaging. |
#82
|
|||
|
|||
donaldjcecil wrote:
Arny Krueger wrote: donaldjcecil wrote: Do you know about SATA implementation on the ASUS A8V-E Deluxe? The chipset listed for the product is VIA K8T890 and VIA VT8237R. http://www.via.com.tw/en/products/ch...series/k8t890/ looks pretty good on paper! ;-) Arny, so based off that diagram you can in your best judgment say that that via chipset doesn't have routing that could bog down the PCI bus? Yes, and also the fact that I've observed the operation of a few dozen Via chipset-based A64 machines that I have so far provided to customers. They do a nice job of exploiting SATA drives. These machines also had SATA RAID controllers on-board which I am exploiting more and more. An A64 machine with mirrored hard drive is a nice combination of speed and robustness. I don't think that it is well-known that mirroring provides about half of the speed advantage of striping in typical use. Given that modern high-capacity 7200 rpm drives while continuing to lag CPU speeds are still getting faster, this yields a nice frisky machine. Given that hard drives (and fans) are the least reliable parts of modern PCs, investing the current cheap hard drive space in mirroring can make a lot of sense. |
#83
|
|||
|
|||
"Edi Zubovic" edi.zubovic[rem wrote in message
... On Mon, 27 Jun 2005 19:52:18 -0400, "Roger W. Norman" wrote: -------------8-------------------- Maxtor Atlas drives are great too. Actually I have two Fujitsu 10k rpm 18 giggers sitting on the floor right now. And a 33 gig 15k rpm 1 gigabit fiberchannel drive. They didn't test any better than my 80 gig ATA Baracudas, but for some reason one gets the impression that they are stronger, if that's even a term one can use with a drive. Robust may be better. But then again, I have/need a lot of data on my computer because of the way I work and the fact that I'm moving into video along with audio for those location recordings. On my last location shoot there was 36 gigs of 16/44.1 track data alone, minus the video transfer data requirement. Once the shoot/audio was edited and compiled it still took two full DVDs. So that's kinda what I'm shooting for. Enough SCSI for work like that will cost about 4 times the money for simple ATA or SATA drives. -- Yes, the cost is an issue here. If the time plays a role, it can pay back as the work is done sooner. For such big files as video, a SCSI RAID0 would be even better. But unfortunately, it takes a much greater costs toll, yes. On the brighter side, SCSI drives are getting faster with each generation and their prices, while being still higher, seems to be more keen compared to those of sveral years ago. Why RAID0 in such cases -- from some test graphs which can be found in the HD Tach 3.0* benchmark program database, I've seen that the transfer rate remains constant all the time. So I think, in RAID0, one disk reads the data as usual, from a to z, but the other disk reads the same data the other way, from z to a. This I think compesates for transfer rate droop due to disk geometry. So this can add to time savings in work too. For instance, I just opened a recording I made yesterday evening in a theatre, it's about 1hr 12min and has about 770 MB. It took some 8 sec. to open it from the first instance and abt. 2,5 sec. to open it again. With a RAID0, it would take slightly less but had I a couple of GB file, it could spare some time depending how much area at the platter are covered by that data. * This is the link, the database comes with the program. http://www.simplisoftware.com/Public...request=HdTach Edi Zubovic, Crikvenica, Croatia I have HD Tach 3 and the SATA raid is terribly inconsistent so I've started research on a replacement Gigabyte mobo that utilizes the nVidia NForce 3 250 Ultra chipset for the SATA connections (well, two of the four). This should be an "in step" upgrade just to accomplish the fix. The next step upgrade would force me to purchase a 939 A64 as well and that translates to about a $500 upgrade. I'm busy with money going out at a fair clip with a new roof, new plumbing and new siding plus the install on a new split mini HVAC unit, so another major computer upgrade is out of the question. I might just go ahead and put my two Fujitsus onto the 29160 and raid them for a while. It's only 36 gigs total, but it's not reasonable to have a studio and no way to record! g Or at least something I can't trust. But speaking of new roofing I must go move my tomato plants before they get crushed. Thanks for the ideas. -- Roger W. Norman SirMusic Studio |
#84
|
|||
|
|||
"Mike Rivers" wrote in message news:znr1119785751k@trad... In article writes: If you are putting a new box together today, the price of 64 bit components is not that much higher than what you would spend for a similarly high end box. Might as well get a Socket 939 MB and a cheap Athlon 64 processor. You can upgrade the processor to a socket 939 compatible dual core Athlon 64 in another 6 months when the prices come down and upgrade to XP 64 or Longhorn So you're suggesting that I might want to buy something that I'll upgrade in six months? You obviously don't know me. I occasionally make a mistake and buy something that breaks in six months, but I don't buy things that I'll have to, or even want to upgrade in six months. What I'm suggesting is that if one was building a new box that they get a motherboard that has a socket configuration that will support future processors. In your context that would mean getting a buggy that would allow you to have it pulled by a different horse at some point in the future when your current mule gets hauled away to the glue factory. I have no problem with people who are happy with less than new equipment, but it would be penny foolish to ignore upgradeability and the possibility of enough raw power and headroom in a computer to multi-task with sonic impunity. I guess we all have different senses of value, but many of us prefer the computer equivalent of a U-87 over a SM-57, and to be quite honest, a difference of a few hundred dollars when building a new computer for professional use is quite meaningless. |
#85
|
|||
|
|||
On Tue, 28 Jun 2005 09:22:18 -0400, "Roger W. Norman"
---------8---------------- I might just go ahead and put my two Fujitsus onto the 29160 and raid them for a while. It's only 36 gigs total, but it's not reasonable to have a studio and no way to record! g Or at least something I can't trust. But speaking of new roofing I must go move my tomato plants before they get crushed. Thanks for the ideas. You're welcome, but-- how you are going to raid the drives with a 29160? I think you'd need a SCSI Raid controller (a bit more $$)? Edi Zubovic, Crikvenica, Croatia |
#86
|
|||
|
|||
Charles Tomaras wrote:
What I'm suggesting is that if one was building a new box that they get a motherboard that has a socket configuration that will support future processors. This seems like a good idea on the face of it, but in practice it is a tad less than optimal. If you buy an early-life cycle motherboard odds are pretty good that it isn't going to be nearly as slick and well-debugged as say a mid-life cycle motherboard for the same CPU. In contrast, CPU's tend to be pretty bug free at any point in their life cycle. |
#87
|
|||
|
|||
Bob Cain wrote:
wrote: There are performance benefits especially when doing 64bit computations on a 64bit processor. Most certainly, but what's in question is the relevance of that to a DAW. I'm certain there are some plugins that can benefit from it. -- Aaron |
#88
|
|||
|
|||
Bob Cain wrote:
Scott Dorsey wrote: Bob Cain wrote: wrote: There are performance benefits especially when doing 64bit computations on a 64bit processor. Most certainly, but what's in question is the relevance of that to a DAW. Right now, it's totally irrelevant. But I bet it turns out to make it easier to design good fast reverbs. Double-precision floats can be a good thing for reverb simulation. I may be wrong but I think that FP units have 64 bit wide data paths even in 32 bit machines. In and of itself, 64 bit is really about integers and addresses. That there is a larger register set in 64 bit mode, which I think is pretty meager in 32 bit x86 architecture, is a definite plus. ....and 64bit machines have 128bit FP datapaths (or more) in some cases. Anybody know what the number of addressable registers is in each? Pretty irrelevant on the Intel parts considering replay and register renaming. -- Aaron |
#89
|
|||
|
|||
Arny Krueger wrote:
In contrast, CPU's tend to be pretty bug free at any point in their life cycle. Now this is funny.... -- Aaron |
#91
|
|||
|
|||
"Edi Zubovic" edi.zubovic[rem wrote in message
... On Tue, 28 Jun 2005 09:22:18 -0400, "Roger W. Norman" ---------8---------------- I might just go ahead and put my two Fujitsus onto the 29160 and raid them for a while. It's only 36 gigs total, but it's not reasonable to have a studio and no way to record! g Or at least something I can't trust. But speaking of new roofing I must go move my tomato plants before they get crushed. Thanks for the ideas. You're welcome, but-- how you are going to raid the drives with a 29160? I think you'd need a SCSI Raid controller (a bit more $$)? Edi Zubovic, Crikvenica, Croatia Actually, (knock wood) I think I fixed the problem. I uninstalled the SATA raid, rebooted, and so far everything is fine. But I'm waiting for the other shoe to fall. In SCSI under XP Pro you don't need a special controller. You can assign SCSI drives (or any drives) under XP Pro as a raid. It's a software raid, that's for certain, but XP Pro does it quite well. I think it was a feature under Win2K, too, but I'm no longer a computer expert. I simply try to do my work with my computers. 15 years ago I was a computer expert. -- Roger W. Norman SirMusic Studio |
#92
|
|||
|
|||
On Tue, 28 Jun 2005 16:38:38 -0400, "Roger W. Norman"
wrote: In SCSI under XP Pro you don't need a special controller. You can assign SCSI drives (or any drives) under XP Pro as a raid. It's a software raid, that's for certain, but XP Pro does it quite well. I think it was a feature under Win2K, too, but I'm no longer a computer expert. I simply try to do my work with my computers. 15 years ago I was a computer expert. Oh! {headsplat} now I remember. Yes it is true, XP does have a software RAID feature. Good luck anyways, Edi Zubovic, Crikvenica, Croatia |
#93
|
|||
|
|||
"Edi Zubovic" edi.zubovic[rem wrote in message
... On Tue, 28 Jun 2005 16:38:38 -0400, "Roger W. Norman" wrote: In SCSI under XP Pro you don't need a special controller. You can assign SCSI drives (or any drives) under XP Pro as a raid. It's a software raid, that's for certain, but XP Pro does it quite well. I think it was a feature under Win2K, too, but I'm no longer a computer expert. I simply try to do my work with my computers. 15 years ago I was a computer expert. Oh! {headsplat} now I remember. Yes it is true, XP does have a software RAID feature. Good luck anyways, Edi Zubovic, Crikvenica, Croatia Things seem to be working fine, but there's always tomorrow! g -- Roger W. Norman SirMusic Studio |
#94
|
|||
|
|||
"Charles Tomaras" wrote in message
... "Mike Rivers" wrote in message news:znr1119785751k@trad... In article writes: If you are putting a new box together today, the price of 64 bit components is not that much higher than what you would spend for a similarly high end box. Might as well get a Socket 939 MB and a cheap Athlon 64 processor. You can upgrade the processor to a socket 939 compatible dual core Athlon 64 in another 6 months when the prices come down and upgrade to XP 64 or Longhorn So you're suggesting that I might want to buy something that I'll upgrade in six months? You obviously don't know me. I occasionally make a mistake and buy something that breaks in six months, but I don't buy things that I'll have to, or even want to upgrade in six months. What I'm suggesting is that if one was building a new box that they get a motherboard that has a socket configuration that will support future processors. In your context that would mean getting a buggy that would allow you to have it pulled by a different horse at some point in the future when your current mule gets hauled away to the glue factory. I have no problem with people who are happy with less than new equipment, but it would be penny foolish to ignore upgradeability and the possibility of enough raw power and headroom in a computer to multi-task with sonic impunity. I guess we all have different senses of value, but many of us prefer the computer equivalent of a U-87 over a SM-57, and to be quite honest, a difference of a few hundred dollars when building a new computer for professional use is quite meaningless. It a quest for the holy grail of computing. One will never achieve it. One can buy the newest and yet still come out shorthanded. It happens. -- Roger W. Norman SirMusic Studio |
#95
|
|||
|
|||
On Tue, 28 Jun 2005 07:14:56 -0400, "Arny Krueger"
wrote: I don't think that it is well-known that mirroring provides about half of the speed advantage of striping in typical use. Given that modern high-capacity 7200 rpm drives while continuing to lag CPU speeds are still getting faster, this yields a nice frisky machine. I sure didn't know that. Very cool; thanks! Chris Hornbeck "I can build you a test that will show either one. Which would you prefer me to demonstrate?" --scott |
#96
|
|||
|
|||
Arny Krueger wrote:
donaldjcecil wrote: Arny Krueger wrote: donaldjcecil wrote: Do you know about SATA implementation on the ASUS A8V-E Deluxe? The chipset listed for the product is VIA K8T890 and VIA VT8237R. http://www.via.com.tw/en/products/ch...series/k8t890/ looks pretty good on paper! ;-) Arny, so based off that diagram you can in your best judgment say that that via chipset doesn't have routing that could bog down the PCI bus? Yes, and also the fact that I've observed the operation of a few dozen Via chipset-based A64 machines that I have so far provided to customers. They do a nice job of exploiting SATA drives. These machines also had SATA RAID controllers on-board which I am exploiting more and more. An A64 machine with mirrored hard drive is a nice combination of speed and robustness. I don't think that it is well-known that mirroring provides about half of the speed advantage of striping in typical use. Given that modern high-capacity 7200 rpm drives while continuing to lag CPU speeds are still getting faster, this yields a nice frisky machine. Given that hard drives (and fans) are the least reliable parts of modern PCs, investing the current cheap hard drive space in mirroring can make a lot of sense. This is also considered Raid 1, is it not? Should I invest in Raid 1, 0, or both?? I understand the motherboard I'm considering does indeed have Raid 1+0, but would that require 3 or more hard drives? |
#97
|
|||
|
|||
donaldjcecil wrote:
Arny Krueger wrote: donaldjcecil wrote: Arny Krueger wrote: donaldjcecil wrote: Do you know about SATA implementation on the ASUS A8V-E Deluxe? The chipset listed for the product is VIA K8T890 and VIA VT8237R. http://www.via.com.tw/en/products/ch...series/k8t890/ looks pretty good on paper! ;-) Arny, so based off that diagram you can in your best judgment say that that via chipset doesn't have routing that could bog down the PCI bus? Yes, and also the fact that I've observed the operation of a few dozen Via chipset-based A64 machines that I have so far provided to customers. They do a nice job of exploiting SATA drives. These machines also had SATA RAID controllers on-board which I am exploiting more and more. An A64 machine with mirrored hard drive is a nice combination of speed and robustness. I don't think that it is well-known that mirroring provides about half of the speed advantage of striping in typical use. Given that modern high-capacity 7200 rpm drives while continuing to lag CPU speeds are still getting faster, this yields a nice frisky machine. Given that hard drives (and fans) are the least reliable parts of modern PCs, investing the current cheap hard drive space in mirroring can make a lot of sense. This is also considered Raid 1, is it not? Should I invest in Raid 1, 0, or both?? I understand the motherboard I'm considering does indeed have Raid 1+0, but would that require 3 or more hard drives? And just to add on to my previous post, I don't know about Raid, just a little bit. I don't understand which will serve a better purpose for audio, 1 or 0. I appreciate your expertise and reponse. |
#98
|
|||
|
|||
|
#99
|
|||
|
|||
"donaldjcecil" wrote in message
news:VGlwe.10255$ro.7762@fed1read02 This is also considered Raid 1, is it not? Raid 0 is striping, Raid 1 is mirroring Should I invest in Raid 1, 0, or both?? Every commercial PC implementation I'm familiar with provides both - you choose. I understand the motherboard I'm considering does indeed have Raid 1+0, but would that require 3 or more hard drives? No, just 2 drives And just to add on to my previous post, I don't know about Raid, just a little bit. I don't understand which will serve a better purpose for audio, 1 or 0. They both have benefits over no RAID at all. Raid 1 provides instantaneous backup for all your data, total failure of one drive will not cause any data to be lost. But, it does this at a cost, you need 2 gigabytes worth of hard drives for every 1 gigabyte of storage space. And, you need to have 2 hard drives. Raid 0 provides an overall speed advantage, but makes data loss twice as probable. |
#100
|
|||
|
|||
Bob Cain wrote:
wrote: ...and 64bit machines have 128bit FP datapaths (or more) in some cases. Does that drop back to 64 in 32 bit mode? That is going to depend on the individual instruction in question - for SIMD type operations (SSE/SSE2/SSE2/Altivec etc) odds are it does not drop down to using only the lower 64bits when running in 32b mode. The question I'm addressing is not whether the new 64 bit machines have a more advanced architecture, that's to be expected, but whether or not an audio app should have any sensitivity to the mode it is compiled for so as to gain performance going to 64 bit mode. Re-compiling will always be the biggest gain, even going between different steppings of the same processor - but even different pieces of silicon sold as the same aren't necessarily the same. Anybody know what the number of addressable registers is in each? Pretty irrelevant on the Intel parts considering replay and register renaming. Doesn't the compiler still map variables to registers for purposes of enhancing performance? The assumption is that the addressable set is faster to use than main memory. Thus the more that are are addressable for the compiler to allocate variables to, the better the performance independant of the factors you name. Wrong? With renaming you can map numerous things to the same register though, and things like renaming and replay allow you to hide the latency of trips to main memory by doing other computations in their stead while you wait for memory. -- Aaron |
Reply |
Thread Tools | |
Display Modes | |
|
|
Similar Threads | ||||
Thread | Forum | |||
Why Windows is Easier than Linux For An End User, Especially for Multimedia work. | Pro Audio |