Yes you're talking about loading times, SSD's sequential performance for reads is going to win. However once an app is loaded (or sampleset etc) then is there still a performance advantage? For video, probably but space is a premium there and a 3-disk raid might be more cost effective. For audio, with large sample libraries certainly. For someone who records bands or largely tracks out other audio material? Possibly not, and if you do a lot of custom loop work & resampling the write load on the drive might cause rather high wear levelling with some of the drives on the market right now (sandforce's for instance.) For a laptop or a Windows install that averages ~30-40GB the boot times with even a 60GB drive make it worthwhile now since it's not a major investment.
I realize I'm far from a normal case in respect to the overall size of the drives that I need (for OS, for some of my Video work, etc) One caveat I suppose is that the larger drives are where things do start to get further down the $/GB price curve. But the way things work now by spreading my workload out across 5 drives allows me to utilize the faster portion of my drives for working tasks (5 is a lot for that) and the middle for general work/storage with the ends of drives for backup & longterm/rarely touched storage. Having 4TB in a single machine for video & audio work is a real bonus, and it's hard to give that up for a few seconds shaved off of apps. Also fwiw an SSD will make me want a backup partition that much more, give the way they tend to fail.
For audio, I get a 'random samples' drive (I have a large collection of 'things' collected over the last 2 decades that are mostly raw audio), a 'sample library' drive (NI stuff and so on), a working projects drive, and my 5th drive is untouched so I suppose should rompler/sampler voicecounts become an issue I can shuffle some work there. When I run Photoshop & Illustrator, they both have dedicated swap drives (1 each), my primary 'project drive', a 'shared' second swap drive and my os/apps drive. The same goes for when I switch to After Effects+Premiere (video) or SoftimageXSI/Blender+Nuke (3d/fx), each app gets a dedicated 'cache' drive and a 'work' drive, plus my OS/apps drive (no spare.)
I'd love to get an SSD or two but my SATA headers are full and anything less than 160-200GB will compromise my workflows. Right now my main 'workflow' issue is just having projects to work on and how I interleave things by scheduling them properly so they don't overlap too much. If anything I've probably put too much thought into my drive setup...
Mixing on SSD
Re: Mixing on SSD
sure , at 200€ the 128Go, it is less interesting than 75€ for 1To....
Re: Mixing on SSD
sorry spacef I edited that! I hit submit instead of preview and almost double posted.
In any case I do hope that the next-gen of MLC nand is here by the time I do my next desktop build. Once that hits I'll also probably be doing a dvd-bay swap on a mac laptop for a pair of SSD's for a video/quartz composer vj machine. Again though for audio use I don't really have issues with my 500GB 7200rpm drive + Ableton Live/Max4Live + Serato ("the bridge" mode with ableton) in terms of drive performance, just cpu for low latency use...
In any case I do hope that the next-gen of MLC nand is here by the time I do my next desktop build. Once that hits I'll also probably be doing a dvd-bay swap on a mac laptop for a pair of SSD's for a video/quartz composer vj machine. Again though for audio use I don't really have issues with my 500GB 7200rpm drive + Ableton Live/Max4Live + Serato ("the bridge" mode with ableton) in terms of drive performance, just cpu for low latency use...
Re: Mixing on SSD
In all honesty its the application that is where the optimizations should be.
Kontakt 4.2 is amazingly fast on my ancient Raptor array ( JBOD ).
I do have Seagate Momentus XL Hybrids that are great for OS+Apps as it learns you routines and then after 8 reboots speeds things significantly.
My 128GB Corsair is fast for apps and OS too, but having 3 x Raptors with the content spanned is as fast as the SSD.
So IMHO NI is making great gains in advancing their sampler app.
I have even dumped old CD's from the 90's like EPershings SFX and voila. I can now rebuild my database.
I have stayed away from East/West PLAY since it's an in house application that seems to be a fine app for massive MacPro's and SSD,s.....etc.etc.
But I have read so many people are struggling to get it record......???
I seriously doubt I could ever dream of using that massive CPU/RAM hog live. If a single track is that troublesome, why bother..?
Anyways, NCW is a new lossless compression technique, and I am using LASS, which is a large String instrument redone with NCW.......THis is a bad mofo.
I can zip through setups, and load and play in realtime....!!!!!
If these realtime features are important to anyone, I suggest if they haven't done so yet, jump on x64 Kontakt 4.2 this summer, and don't waste your money anymore on these giant poorly coded Orchestral behemoths.
I actually have caused crashes just to see how fast my 6GB template takes to reload in a live scenario, and it takes 3 seconds.....??
I think these boys have an excellent memory server trick or two and can never see using anything other than Kontakt live again.............EVER.
I remember when I first tried Generator back in 98' and laughed at how weak it was, I even laughed at Reaktor and yes Kontakt sucked too, way too thin and unstable until Kontakt 3.0.
But just in the last year and a half these guys have taken an indoor sample playback app, and turned it into a live hunting dog.
They have my cash for a long time now.
Now all we have to do is get kids off of these button pushing apps and teach them Piano, Guitar, Drums and horns again.
THere's still hope for humanity...
Kontakt 4.2 is amazingly fast on my ancient Raptor array ( JBOD ).
I do have Seagate Momentus XL Hybrids that are great for OS+Apps as it learns you routines and then after 8 reboots speeds things significantly.
My 128GB Corsair is fast for apps and OS too, but having 3 x Raptors with the content spanned is as fast as the SSD.
So IMHO NI is making great gains in advancing their sampler app.
I have even dumped old CD's from the 90's like EPershings SFX and voila. I can now rebuild my database.
I have stayed away from East/West PLAY since it's an in house application that seems to be a fine app for massive MacPro's and SSD,s.....etc.etc.
But I have read so many people are struggling to get it record......???
I seriously doubt I could ever dream of using that massive CPU/RAM hog live. If a single track is that troublesome, why bother..?
Anyways, NCW is a new lossless compression technique, and I am using LASS, which is a large String instrument redone with NCW.......THis is a bad mofo.
I can zip through setups, and load and play in realtime....!!!!!
If these realtime features are important to anyone, I suggest if they haven't done so yet, jump on x64 Kontakt 4.2 this summer, and don't waste your money anymore on these giant poorly coded Orchestral behemoths.
I actually have caused crashes just to see how fast my 6GB template takes to reload in a live scenario, and it takes 3 seconds.....??
I think these boys have an excellent memory server trick or two and can never see using anything other than Kontakt live again.............EVER.
I remember when I first tried Generator back in 98' and laughed at how weak it was, I even laughed at Reaktor and yes Kontakt sucked too, way too thin and unstable until Kontakt 3.0.
But just in the last year and a half these guys have taken an indoor sample playback app, and turned it into a live hunting dog.
They have my cash for a long time now.
Now all we have to do is get kids off of these button pushing apps and teach them Piano, Guitar, Drums and horns again.
THere's still hope for humanity...
Re: Mixing on SSD
hi greg,siriusbliss wrote:See for programs.
Esata for projects
Adjust buffer settings in host.
Turn off all background processes
Add more RAM
I've had most success with this approach.
Greg
tell me more about buffer settings in host. whats your sequencer and buffer?
thanks
Re: Mixing on SSD
Hi all,
I found that a hard drive
Samsung SpinPoint F3 - HD103SJ - 1 To 7200 RPM 32 Mo Serial ATA II
is way faster than any of my caviar black or blue
it's like 80/100 MB/s (Caviar Black 1 year old/64MB cache, blue=100/120 32MB cache....)
vs 140/120 MB/s (Samsung, ref above, two weeks old (it's new)
Measured with HD Tune http://www.hdtune.com/
apparently, cache is not something that improves performances automatically in all situations....
I wish i could replace all my WDs for Samsung drives, but well, later on.....
ps: we also measured the SSD (OCZ Agility 128GB, anouced as more than 250 mb/s transfer rates) of my mate and it gives very bad results with HD Tune (but I don't think HD Tune is able to measure SSDs): around 35 mb/s only on ssd.... i guess that's why it is not advised for recording audio....
I found that a hard drive
Samsung SpinPoint F3 - HD103SJ - 1 To 7200 RPM 32 Mo Serial ATA II
is way faster than any of my caviar black or blue
it's like 80/100 MB/s (Caviar Black 1 year old/64MB cache, blue=100/120 32MB cache....)
vs 140/120 MB/s (Samsung, ref above, two weeks old (it's new)
Measured with HD Tune http://www.hdtune.com/
apparently, cache is not something that improves performances automatically in all situations....
I wish i could replace all my WDs for Samsung drives, but well, later on.....
ps: we also measured the SSD (OCZ Agility 128GB, anouced as more than 250 mb/s transfer rates) of my mate and it gives very bad results with HD Tune (but I don't think HD Tune is able to measure SSDs): around 35 mb/s only on ssd.... i guess that's why it is not advised for recording audio....
Re: Mixing on SSD
Random iops (i/o operations) is where SSD's fall down, but that should be changing this year.
-
jksuperstar
- Posts: 1638
- Joined: Mon Nov 15, 2010 12:57 pm
Re: Mixing on SSD
Really? I've always had great experience with Random I/O on SSD, since it's more like RAM, than linear access like a hard drive waiting for the next spin to complete and be able to read the data (isn't that why people like 7.5k or 10k drives over 5400?).
Re: Mixing on SSD
Sorry you're right, my response wasn't very clear to what I was thinking at the time, I must have been distracted and I think a good deal of text is missing....? I'll try to reform what I was thinking in my fragmented reply above...
As you scale up the width of the drive the write performance is mitigated both by increasing parallelism and the ability to interleave writes (and of course modern garbage collection + trim eliminates the old write cycle hit). So with drives large enough (typically over 120GB or so) the performance is of course peak for the class (SSDs) and thus trounces magnetic/spinning HD's. But that of course isn't the performance of the nand chips themselves, it's the performance of the specific configuration of parallel chips and possibly dual controllers (essentially raid right on the ssd) overcoming the limitations of MLC nand.
So what I was really thinking of above was that we're right on the cusp of the point where the nand chips are small & fast enough that even the lower priced drives will be large & fast enough to make choosing them a clear winner. Of course the failure rates are almost halving each generation so that more & more 'spare' space must be put aside and error correction is going to become a prominant feature even in non-sandforce drives, and new chipsets come with unknown risks.
But SSD's to me as a tech just feel like early PATA/SCSI cost differences, and I dealt with that through much of the previous decade. For several years now it's been nice to be rid of the need of the cost of a scsi controller + the appropriate disks. So I'm hoping the leap that's coming this year will push the market of SSD's to the point where volume & the performance of even mainstream parts pushes the sweet spot for performance under $200 for SSD's, but as of yet Intel & OCZ/OWC drives seem to be driving the market by increasing performance each generation without the expected price drops.
Basically I think that drives need to be ~$150 (or less) in their 'full' form, not some half-width consumer oriented form that's affordable (for netbooks etc) with 'real' drives costing as much as a cpu that will probably outlast it. The size of the drive will of course still scale with cost, but as things stand now you're (imo) getting 'half' the performance each time you halve the pricepoint of your SSD purchase. Past 100GB, the X-25m G2's, C300's & Vertex 2's are of course good enough now but compare the performance differences between the $100-150 drives, the $250-350 drives and the ones that run $400-600 and you'll see the performance is scaling with the 'width' of the drives as the capacity does too. Then look at the half-width drives that fall under 80GB (30/40GB drives especially) and their performance, and it's clear that you've still got to consider the controller+chip combination when making a purchase.
Now that's one issue (cost as a barrier to entry vs. performance) but another issue is how your workload is tuned to the SSD. If I'm going to spend $300-400 on a drive (the higher ones are unnecessary to me) then I want to make sure that I'm not thrashing the drive unnecessarily (affecting its lifespan especially as we go down in nand gate size.) For instance in sequential & random writes, newer drives are doing much better at 4k because this fits with the size of the nand chips, and because WIn7 & OSX 10.6 are prepared for 4k data blocks now as well (WD has the first magnetic HD to use 4k sectors now too, and the others are following suit.)
Sorry this was somewhat rambling. I know the point I was initially trying to make but since about 3x as much briandump is occuring I think it's more obscured. But hopefully what I said is more accurate in the context of said braindump...
As you scale up the width of the drive the write performance is mitigated both by increasing parallelism and the ability to interleave writes (and of course modern garbage collection + trim eliminates the old write cycle hit). So with drives large enough (typically over 120GB or so) the performance is of course peak for the class (SSDs) and thus trounces magnetic/spinning HD's. But that of course isn't the performance of the nand chips themselves, it's the performance of the specific configuration of parallel chips and possibly dual controllers (essentially raid right on the ssd) overcoming the limitations of MLC nand.
So what I was really thinking of above was that we're right on the cusp of the point where the nand chips are small & fast enough that even the lower priced drives will be large & fast enough to make choosing them a clear winner. Of course the failure rates are almost halving each generation so that more & more 'spare' space must be put aside and error correction is going to become a prominant feature even in non-sandforce drives, and new chipsets come with unknown risks.
But SSD's to me as a tech just feel like early PATA/SCSI cost differences, and I dealt with that through much of the previous decade. For several years now it's been nice to be rid of the need of the cost of a scsi controller + the appropriate disks. So I'm hoping the leap that's coming this year will push the market of SSD's to the point where volume & the performance of even mainstream parts pushes the sweet spot for performance under $200 for SSD's, but as of yet Intel & OCZ/OWC drives seem to be driving the market by increasing performance each generation without the expected price drops.
Basically I think that drives need to be ~$150 (or less) in their 'full' form, not some half-width consumer oriented form that's affordable (for netbooks etc) with 'real' drives costing as much as a cpu that will probably outlast it. The size of the drive will of course still scale with cost, but as things stand now you're (imo) getting 'half' the performance each time you halve the pricepoint of your SSD purchase. Past 100GB, the X-25m G2's, C300's & Vertex 2's are of course good enough now but compare the performance differences between the $100-150 drives, the $250-350 drives and the ones that run $400-600 and you'll see the performance is scaling with the 'width' of the drives as the capacity does too. Then look at the half-width drives that fall under 80GB (30/40GB drives especially) and their performance, and it's clear that you've still got to consider the controller+chip combination when making a purchase.
Now that's one issue (cost as a barrier to entry vs. performance) but another issue is how your workload is tuned to the SSD. If I'm going to spend $300-400 on a drive (the higher ones are unnecessary to me) then I want to make sure that I'm not thrashing the drive unnecessarily (affecting its lifespan especially as we go down in nand gate size.) For instance in sequential & random writes, newer drives are doing much better at 4k because this fits with the size of the nand chips, and because WIn7 & OSX 10.6 are prepared for 4k data blocks now as well (WD has the first magnetic HD to use 4k sectors now too, and the others are following suit.)
Sorry this was somewhat rambling. I know the point I was initially trying to make but since about 3x as much briandump is occuring I think it's more obscured. But hopefully what I said is more accurate in the context of said braindump...
Re: Mixing on SSD
Here's a good illustration of what I was referring to above, on the first page of anandtech's Vertex3 240GB review he explains at length why the larger drives with 'full' capacity drives when they are able to take advantage of the high level of parallelism with a drive that is fully populated with large chips, ie interleave the reads & writes:
http://www.anandtech.com/show/4316/ocz- ... 0gb-review
Sadly this drive is still >$500 here...and even the 'half width' 126GB version is still ~$300.
http://www.anandtech.com/show/4316/ocz- ... 0gb-review
Sadly this drive is still >$500 here...and even the 'half width' 126GB version is still ~$300.