|
How much cpu power do i need? (pg. 3)
|
View this Thread in Original format
| DigiNut |
| quote: | Originally posted by Subtle
Soon it is not going to matter which CPU we have, we need to have fast RAM and Hard Drives to notice the difference. |
What you seem to be suggesting would require a breakthrough in computing and manufacturing. DRAM is still several orders of magnitude slower than CPU, and magnetic disk is still several orders of magnitude slower than DRAM. But for the most part, you don't notice the memory speed at all because it's fast enough, and disk bottlenecks are mostly eliminated through smart caching.
Mostly, the bottleneck is just inefficient software. Trust me - I have seen some doozies, operations that take 10 minutes that could easily be performed in less than 1 second using the exact same architecture.
I get what you're trying to say, really, but it's still wrong. Removing the CPU bottleneck doesn't suddenly mean the bottleneck is now somewhere else. It does if the system is at full load, but in the case of production it's nowhere near full load in terms of memory/disk/network bandwidth. You get yourself a fast quad-core CPU and a decent ASIO sound board and it frankly doesn't matter whether you're using DDR2 or DDR3 or have a 7200 RPM or 10K RPM or solid state disk - the experience is going to be exactly the same.
The reason I am going on about this - just so you don't think I'm merely picking on you - is that it seems to imply that if you get a fast CPU, you also need to pick up the highest-end memory and hard disks in order to get the benefits of that, and for the vast majority of us that is not the case. In my many years of experience I have noticed a minuscule difference if any between a system running the fastest OCZ or Crucial memory and the same system running the generic high-latency stuff. It's a waste of money. |
|
|
| DJ Robby Rox |
| quote: | Originally posted by 4everX
depends on how many plug-in you use, i've a [email protected], 4gb Ram@800mhz and i use 12 or 13 istance of virtual instrument plus a lot of effects (eq, compressor delay...)
my CPU is often over 50% about 60-70% |
Thats ideally how many I'd like to use.
I get up to around 6, my CPU starts choking so I dive into my massive soundfont collection.. than my RAM starts choking.
Than I try layering synths, thickening things up, adding fx, and my computer aint having it.
I am a CPU whore because I use a lot of ghost sounds. Like I'll make a pad, than open up 2 more z3tas to thicken it up. |
|
|
| Subtle |
| quote: | Originally posted by DigiNut
[FONT=Tahoma][COLOR=#99CCEE]
I get what you're trying to say, really, but it's still wrong. Removing the CPU bottleneck doesn't suddenly mean the bottleneck is now somewhere else. It does if the system is at full load, but in the case of production it's nowhere near full load in terms of memory/disk/network bandwidth. You get yourself a fast quad-core CPU and a decent ASIO sound board and it frankly doesn't matter whether you're using DDR2 or DDR3 or have a 7200 RPM or 10K RPM or solid state disk - the experience is going to be exactly the same.
|
The only thing in my current system that i would like to have faster is Omnisphere, the loading of samples are sluggish.
I figured this could be solved with using ultrafast HD like SSD and fast RAM plus an OS like Windows 7 with all that 64 bit stuff.
But if you are saying that it wont make a great difference, then i might rethink my future upgrades. |
|
|
| DigiNut |
SSDs are actually slower than ordinary mechanical hard drives when it comes to streaming large amounts of data or loading them into memory. They're just a lot better for random access, which is why some people are using them as boot drives and why Vista uses flash drives for ReadyBoost (swap file).
If you're loading thousands of small samples - yes, an SSD might make it faster. If you're loading two or three huge samples, you won't see any improvement; it might even get worse.
If you're really getting a lot of disk lag, try (a) putting all your samples on their own drive, and (b) setting up a RAID 0 (striped) array. Theoretically it should take half the time, although the typical onboard SATA RAID solutions are pretty inefficient, so you might only get a 50% increase or so.
As for faster RAM - don't bother, you won't see a difference.
Also, I should mention that depending on your system, the amount of RAM may make a difference; if you exceed the available physical memory then you'll start paging out to disk. This is bound to happen if you're loading enormous sample libraries of orchestras and such. Of course, this is not unique to production, and unless you're running a 64-bit OS, you're limited to about 3 GB anyway. |
|
|
| echosystm |
| quote: | Originally posted by DigiNut
SSDs are actually slower than ordinary mechanical hard drives when it comes to streaming large amounts of data or loading them into memory. They're just a lot better for random access, which is why some people are using them as boot drives and why Vista uses flash drives for ReadyBoost (swap file).
If you're loading thousands of small samples - yes, an SSD might make it faster. If you're loading two or three huge samples, you won't see any improvement; it might even get worse.
|
This all wrong mate. Modern SSDs easily out perform hard drives on read performance, regardless of IO size and whether it is sequential or random. As you will see from the graphs below, we're talking a huge performance increase with an OCZ Vertex, over the fastest SATA HDD available today - a Velociraptor. On random reads, Velociraptors in RAID still can't keep up. Considering that the typical user would be requiring about 80% random and 20% sequential, SSDs blow HDDs away on everything other than price lol.
The problem with most SSDs is that they are MLC, so you get sub-block fragmentation and the performance degrades over time. This is reset by formatting, but obviously that is an annoying thing to do every few months. Even with sub-block fragmentation, SSDs will out perform a HDD on read performance. I can't remember, but I think read performance isn't affected as much as write performance.
The technology is a lot better than HDD, it's just too costly at the moment. When non-MLC SSDs are affordable, there will be no reason to buy HDDs, even if they stay at current performance levels. |
|
|
| Stef |
| quote: | Originally posted by echosystm
serious? what daw?
i'm using cubase and windows xp. i keep my computer pretty clean so theres no bull in the background. at 5.8ms i have no problem running a song like...
2x legacy cell
1x zebra2
1x halion one
2x guru
3-4 compressors, eqs on most channels (usually glisseq), limiter on master, an amp simulator, a distortion plugin or two, glitch/camelphat or some other multieffect.
something like that wouldn't go much above 50% for me. if i tried to do the same thing in ableton though, i wouldn't even get halfway. |
But this just leads me to ask if i can hear one of your tracks? |
|
|
| echosystm |
| quote: | Originally posted by Stef
But this just leads me to ask if i can hear one of your tracks? |
sure man, pm me your msn
keep in mind though, i don't make trance. house is a lot less vsti-intense. |
|
|
| Stef |
| quote: | Originally posted by echosystm
sure man, pm me your msn
keep in mind though, i don't make trance. house is a lot less vsti-intense. |
Sweet man, sent a pm your way. |
|
|
| noicuc |
All of it.
I use approx 10+ vst ,
(5 nexus , few gladiators and there..)
on a 1.66 ghz centrino duo on fruity loops.
on 263 ms... well you can imagine. |
|
|
| DigiNut |
| quote: | Originally posted by echosystm
|
The figures you're showing don't jibe with the more mainstream benchmarks:
 
 
Various different SSDs outperform a regular VelociRaptor in certain categories, but almost none of them are consistently better and they're almost always inferior to a RAID 0 VR config.
Keep in mind, also, that these 250 GB SSDs are brand-spankin' new. Almost anything built in 2008 is going to have performance roughly equivalent to that SuperTalent drive, i.e. major suck.
EDIT: Now I see the watermark and realize you've taken your results from the same site, haha. Except you cherry-picked the ONE benchmark that actually supported your case, which comes from some obscure and outdated tool that nobody uses anymore. Win!
| quote: | | Considering that the typical user would be requiring about 80% random and 20% sequential, SSDs blow HDDs away on everything other than price lol. |
The typical user, yes. Sample-heavy production, absolutely not!
| quote: | | The technology is a lot better than HDD |
OK, that's quite a grand claim to make. Even if the performance were consistently better (scant evidence), cost is part of the "better" formula. By this logic we could say that SRAM is "better" than DRAM since it's faster and doesn't have to be refreshed, but it's so much more costly to produce and for such a marginal end-user gain that it's only ever used... um, nowhere. |
|
|
| echosystm |
| quote: | Originally posted by DigiNut
EDIT: Now I see the watermark and realize you've taken your results from the same site, haha. Except you cherry-picked the ONE benchmark that actually supported your case, which comes from some obscure and outdated tool that nobody uses anymore. Win!
|
No, I didn't cherry pick anything. I never claimed that write performance was better, because we weren't talking about write performance, we were talking about read performance of samples ;)
The first two benchmarks you have posted are off topic. Likewise, burst speeds don't mean anything in real world use. For the sake of arguing... The reason the RAID config has such a high burst speed is because of the RAID controllers extra cache. If you hooked up an SSD to a RAID controller, you would get the same result. Anyhow, the two decent non-RAIDed SSDs are almost at the same burst speed as the Velociraptor.
This image here shows the exact same thing that my previous post does - good SSDs pwn hard drives on reads. ONE decent SSD easily outstrips the RAID configuration. Anyhow, this benchmark you have shown is old. The OCZ Vertex and Summit SSDs are not on them - these are the first cost effective SSDs to offer good performance. Not having these in the comparison is silly. Considering that Vertex is already out and Summit will be out in a few weeks, the comment below is negated. A few weeks later, something better will come out yet again - SSDs are moving bloody quick at the moment.
| quote: | Originally posted by DigiNut
Keep in mind, also, that these 250 GB SSDs are brand-spankin' new. Almost anything built in 2008 is going to have performance roughly equivalent to that SuperTalent drive, i.e. major suck.
|
The technology is superior, you can't debate that. We are comparing the best hard drive from over 2 decades of development to the crappiest SSDs from only a few years development. It's only a matter of time until they are cheap as piss. Lack of economies of scale are the biggest thing inflating SSD costs. |
|
|
| echosystm |
Here you go:
Clearly, SSDs wipe the floor with Velociraptors, even on write speeds. Once you throw in the extra benefit of having Windows and your applications load like lighting, it's obvious which is better. SSDs are rapidly increasing in speed and decreasing in cost, by the week. Hard drives are stagnating. Let's not continue this silly debate any further.
Here's a few more benchmarks, just so you get it...
 |
|
|
|
|
